For on-site generation, I recommend only using models designed for WAN2.1 or earlier. It does not appear that Civitai can get 2.2 models to work. You can use the Universal model with both high and low noise steps, but if you want the best quality, I included both versions as well.
This LoRA is trained on conversational language. You can describe your scene in plain English or Chinese, and it should adhere to your prompt well. Below are some inspirations for you if you need it.
Positions:
On her back
Kneeling
Perspectives:
POV
Enhancement phrases:
She bounces her body up and down
She bounces her breasts up and down with her hands
All showcase videos have the workflow metadata embedded. You can download the video and drag it into ComfyUI to replicate them with their prompts, seeds, and settings.
Description
Trained on the 1.3B parameter WAN model.
FAQ
Comments (31)
What are your thoughts on WAN? I'm checking out SkyReels right now for the img2vid . Have you tried img2vid on WAN?
the training is avaliable for 14B?
Works great with image 2 vid. thanks!
I2V works best if the whole dick is visible in the first frame!
I’ve already prepared a fully automated workflow for embedding a dick into images and positioning it near a person’s mouth.
@zakakki can you share?
Great LoRA, thanks!
Really like this, I'm just using it get WAN to even make good looking tits regardless of titfuck or not, thanks
When I try to use load lora and use the keyword titfuck I’m not getting anything with wan 2.1 close to it what base model are you using or am I doing the Lora wrong
@Jaynen Are you using the correct LoRA? There are 3 LoRAs here.
@ComfyTinker you are right i think i was using the huanyuan version some loras seem to think they work for both. Do i need to use a load lora node or should the keywords themselves be enough
"HunYuan is the best model for NSFW content at the moment." I think it is if someone uses well-trained loras like yours. Otherwise it can be really hit or miss with character loras. Hope you do more great loras like this one.
I will note that the prompting you suggest in your description usually isn't sufficient. I get consistent results from this one, based on the videos in your showcase:
"She is fully nude and kneeling between the naked viewer's legs. She is holding her breasts with her hands. She looks at the viewer with a playful expression. Titfuck. She slowly titfucks the viewer. She slides her breasts up and down his dick in long slow strokes."
Yeah the prompting is very generic because you can have her titfuck a dildo, side views, have clothes on/off, etc.
I wanted to keep the example prompt simple so that people can add to it to influence in their desired direction.
what IS the best NFSW video generator in your opinion?
@bm3011838 Well the word used was "content" not "video". For video, it is definitely HunYuan from what I can tell. For general purpose prompt adherence, Pony is still the king but it is also mostly garbage at realistic models.
So if you want realistic and/or video Hunyuan is the only way. But if you want maximal NSFW prompt adherence and are into 3D rendered and cartoon girls, Pony is the only way. See my posts for examples. It would be nice if we could get them all in one package but we aren't there yet.
@Clocksmith wan2.1 nsfw？
@pa5min From what I've heard about WAN it can't do genitals.
"Unable to find workflow in ..." - I've tried to download any of the videos, and then drag it into ComfyUI only to get this error message. Regular local ComfyUI.
How do I fix this?
Nice, lots of Wan loras coming too!
How long did the training take?
I have the number of steps/epochs in the details section.
In total, a little over 18000 steps. This was around 30 hours of compute time, and probably another 10 hours of cropping, captioning, and curating.
Amazing LoRA
Works great with I2V
How did you do it? I can't get it to work with I2V
What strength is this tend to be best at for Cartoons and real life?
Should be 1.0 for all. Most of the cartoon/anime style seems to be done using image 2 video (which this LoRA works)
10/10, tits/tits
Can you help me? Why does the character's face change and not resemble the one in the reference image when I use this LoRA? However, this issue does not occur when I use Wan2.1 14B Titty Drop Lora. I have already set the LoRA strength to 0.6, but the character's face still turns into someone else.
Parameters: 480x480, step 20, 5 seconds.
Model: official wan2.1_i2v_480p_14B_fp16.safetensors and other official models.
Scheduler: sgm_uniform, Sampler: dpmpp_2m.
This is also the case for me. Titty drop was trained on I2v, this here was probably trained on T2V.
where do I need this put in?
You load it onto the model using a Lora Loader in ComfyUI.
The boobs 🤭
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.